Shankarblr/Qwen2.5-1.5B-TechWriter-LoRA

Qwen2.5-1.5B TechWriter LoRA adapter

This is not a standalone model.

It is the PEFT LoRA adapter trained with QLoRA on Qwen/Qwen2.5-1.5B-Instruct for semiconductor / data-center interconnect technical writing.

For drop-in inference, use the merged repo:

Shankarblr/Qwen2.5-1.5B-TechWriter-Instruct

Use this repo to resume training, attach the adapter on the frozen 4-bit base, or keep a small artifact.

Load the adapter

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from peft import PeftModel

BASE = "Qwen/Qwen2.5-1.5B-Instruct"
ADAPTER = "Shankarblr/Qwen2.5-1.5B-TechWriter-LoRA"

tok = AutoTokenizer.from_pretrained(ADAPTER)
bnb = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_use_double_quant=True,
    bnb_4bit_compute_dtype=torch.float16,
)
base = AutoModelForCausalLM.from_pretrained(BASE, quantization_config=bnb, device_map="auto")
model = PeftModel.from_pretrained(base, ADAPTER)
model.eval()

Merge if you want a single folder like the inference repo:

merged = model.merge_and_unload()
merged.save_pretrained("./qwen-techwriter-merged")
tok.save_pretrained("./qwen-techwriter-merged")

Training recap

Same run as the merged card:

• Data: private semiconductor technical-writing ChatML mix (6,765 rows, 90/10)
• LoRA r=16, alpha=32 on q/k/v/o/gate/up/down_proj
• 3 epochs, 1,143 steps, ~1 h 30 min
• Eval epoch 3: loss 0.1404, mean token accuracy 0.9488

Upload only adapter_config.json, adapter_model.safetensors, tokenizer files, and this README. Leave out checkpoint-* and optimizer states.

License

Apache 2.0. Unofficial style model; not affiliated with any semiconductor vendor.